Abstract
Recent research suggests that processing in visual search begins with an accumulation of evidence across all locations in the display. This accumulation towards a non-target threshold involves the computation of a contrast signal between each item and the target template. In efficient search tasks, this results in a logarithmic increase in reaction times (RTs) as a function of set size. On the other hand, visual search in scenes can benefit from contextual information by limiting selective attention to target-consistent regions. For example, when looking for a mug, observers limit attention to target-consistent regions (e.g. countertops) while ignoring target-inconsistent regions (e.g. ceilings). Is the evidence accumulation process mandatory for the entire scene in efficient search, or can contextual information exclude target-inconsistent regions from processing? In a series of experiments, we manipulated set size in an efficient search task in scenes. Observers searched for a uniquely-colored target item that always appeared in a target-consistent region (e.g. fish in the sea and not the sky. Consistent with previous studies, we found that the evidence accumulation process is restricted to the set size of the target-consistent region; set size of the target-consistent region did not contribute to RTs. In a subsequent experiment, we changed the predictability of the target-consistent region by making its location random instead of always appearing in a specific area of the display. Thus, observers had to locate the target-consistent region instead of relying on previous experience. Interestingly, all distractors now contributed to RTs. However, observers were not ignoring scene context: logarithmic slopes for distractors in the target-inconsistent region were shallower, indicating that these distractors were rejected more quickly. Thus, evidence accumulation in efficient search is mandatory and takes places across the entire display when scene context is unpredictable but can be speeded up for distractors in target-inconsistent regions.